nfsdr: Nonparametric false simultaneous discovery rate control

Description Usage Arguments Value Examples

Description

Given D sequences of test statistics, returns the optimal square rejection that identifies the largest number of simultaneous signals while controlling the false discovery rate. Assumes a common threshold for each sequence.

Usage

1
nfsdr(T, alpha, rho = 0, m = 5000, rescale = TRUE)

Arguments

T

n x D matrix of test statistics that are stochastically larger under the null, where n is the number of features and D is the numberof sequences of test statistics

alpha

nominal false simultaneous discovery rate

rho

regularization parameter to guarantee asymptotic control of the false discovery rate; should be a small positive value, but rho = 0 works well in most simulations

m

search for the optimal threshold up to only the mth largest unique value of T; can speed up computation

rescale

apply rank transformation to the test statistics within each sequence such that they are of comparable scales

Value

indices of the features delcared to be simultaneous signals

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
## generate paired test statistics
p <- 10^6; ## total number of pairs
X <- c(rep(0,p-30),rep(1,10),rep(2,10),rep(3,10));
## X=0: no signal in either sequence of tests
## X=1: signal in sequence 1 only
## X=2: signal in sequence 2 only
## X=3: simultaneous signal
set.seed(1);
Z1 <- rnorm(p,0,1); Z1[X==1|X==3] <- rnorm(20,3,1);
Z2 <- rnorm(p,0,1); Z2[X==2|X==3] <- rnorm(20,4,1);
T <- cbind(Z1^2, Z2^2);
## rejected simultaneous signals
nfsdr(T, 0.05)

ssa documentation built on May 1, 2019, 10:27 p.m.

Related to nfsdr in ssa...